Computation over Gaussian Multiple-Access Channels

We consider the problem of computing the sum of independent Gaussian sources over a Gaussian multiple-access channel (MAC) with respect to a mean-squared error criterion. When the source and channel bandwidths are equal, the best separation-based solution to this problem performs exponentially worse in a distortion sense compared to the optimal solution: uncoded transmission. In this paper, we develop lattice codes for exploiting the structure of the Gaussian MAC when there are more channel uses than source symbols. We also demonstrate the usefulness of these codes in determining the multicast capacity of a simple AWGN network.